Hugging Face has published a research report titled nanoMuse: An Open-Source Personal Agent for Every Device You Own, introducing an open-source personal agent designed to operate across all of a user's devices.
The report frames the personal agent as a distinct category of software rather than a model. It contrasts earlier assistants, which it says answered and waited, and 2023-era agents, which performed a task and stopped, with a persistent agent that acts on a person's accounts and devices, remembers them across weeks, speaks first when it is worth it, and answers for what it did.
According to the report, Meta's Muse, demonstrated in September 2026, showed such an agent for one person — with accounts, devices, memory and a lasting conversation — but remained closed, hosted in a vendor's cloud and confined to one country. The report says it reads how Muse is built from Meta's public record and a copy of its production prompt, marking each statement by its source, and defines the personal agent through five questions and three horizons.
nanoMuse is presented as the open counterpart under the GPL-3.0 license: one agent on every device a person owns, with hands on the phone's screen and the computer's. The devices share one conversation over a relay anyone can run, every action passes through a Sentinel with scoped approvals, memory is stored as Markdown files the person can read, and the model is the user's choice. The report notes that its size and cost are given as estimates.
The roadmap sets out what remains open: memory with provenance, an evaluation suite for the hands, and an open model for them. Code, a web demo and documentation are linked from the paper.